LAUSR

In Response: We thank Drs Li and Zheng for showing interest in our article. Our main findings were that pre-intravenous thrombolysis (IVT) transcranial doppler (TCD) delayed IVT infusion by a median of 14 minutes. The first of our secondary analyses compared outcomes in patients who had received pre-IVT TCD and those who had not. This stemmed from a hypothesis that an increase in door-to-needle time caused by TCD might lead to worse outcomes. In unadjusted analysis, there were no significant differences between the groups on any outcomes, including SICH, as well as modified Rankin Scale (mRS) score 0 to 1, mRS 0 to 2, and death. Adjustment for baseline differences showed no signal of worse outcomes associated with TCD—on the contrary, there was a borderline significant association with modestly better odds for mRS 0 to 1 and mRS 0 to 2 at 3 months. Furthermore, we saw similar findings on patients who had undergone TCD during IVT infusion, avoiding the preinfusion delay. We had robust findings on the safety of TCD performed in both settings. At the same time, we emphasized the need for caution in interpreting signals of potential benefit. Part of this caution was elaborated as possible impact of beneficial local practice traditions associated with the use of TCD in selected cases. Such center effects and potential country effects can be handled using multilevel modeling, as Drs Li and Zheng correctly point out. We initially deemed this as sufficiently accounted for by the fact that all patient groups, including control groups unexamined with TCD, had been treated at TCD-performing centers, likely under similar diagnostic and treatment traditions associated with TCD use. We have reanalyzed our data in keeping with Drs Li and Zheng’s recommendation. For each outcome, we estimated a 3-level mixed-effect logistic regression model. The nested levels were country, center, and individual. The models included a normal random intercept for country and 1 for center. The set of independent variables used in each model are described in the article. We calculated intraclass correlations for within-center and with-country correlation. The intraclass correlations were slight, values ranging from 0,00 to 0,06 for all outcomes in all analyses. This implies that the logistic regression in the article, which assumed independence from center and country effect, was not far off. Still, in the multilevel models, the point estimates, although similar, were slightly lower for mRS 0 to 2 and mRS 0 to 1, with larger P values. The adjusted odds ratio for mRS 0–2 in Table 2 became 1.09; P=0.30 (from originally 1.15; P=0.043), and the adjusted odds ratio for mRS 0 to 1 in Table III in the online-only Data Supplement became 1.17; P=0.16 (from originally, 1.27; P=0.012). This can partly be accounted for by loss of power because of greater number of parameters in the multilevel models. However, association of during-IVT TCD with decreased mortality by 3 months remained significant after multilevel modeling, adjusted odds ratio, 0.71; P=0.025 (from originally, 0.73; P=0.022). Issues of multicollinearity had already been accounted for in our analyses in the article. Collinearity was seen between onsetto-treatment time and door-to-needle time (Table 2 in the article), which have a perfect correlation. They were not entered into the multivariate model together, but only one at a time, with no changes to the results. This was also seen in systolic and diastolic blood pressure, (Pearson coefficient, 0.5), where the same approach was taken in multivariate modeling. Prior hypertension was only weakly correlated with baseline systolic and diastolic blood pressure (Pearson coefficient, −0.2 and −0.1, respectively). In analyses pertaining to Table III in the online-only Data Supplement, prior hypertension was included in both models interchangeably, including baseline systolic blood pressure and diastolic blood pressure. We conclude that although center and country variance was low, the results following inclusion of center and country effect in the multivariate model confirmed the need to cautiously interpret associations of TCD use with improved outcomes. Our main conclusion, however, still holds that TCD examination, if performed, should be done during IVT infusion to avoid treatment delay.